MongoDB aggregates over row collections of predefined size - mongodb

Question

is it possible in MongoDB to perform aggregations over a collection of predefined number of rows, rather than grouping by. For example I want to calculate the average for every 1000 rows, instead of grouping by a certain column.
A smaller example would be the table below, I would like to calculate the Average rating of every 4 consecutive rows:
So my restult should like somthing like this:
Below is the input data in JSON:
[{"ItemName":"Item1","Rating":4},
{"ItemName":"Item2","Rating":4},
{"ItemName":"Item2","Rating":4},
{"ItemName":"Item3","Rating":2},
{"ItemName":"Item4","Rating":5},
{"ItemName":"Item5","Rating":4},
{"ItemName":"Item6","Rating":2},
{"ItemName":"Item7","Rating":4},
{"ItemName":"Item8","Rating":1},
{"ItemName":"Item9","Rating":4},
{"ItemName":"Item10","Rating":3},
{"ItemName":"Item11","Rating":2},
{"ItemName":"Item12","Rating":2}]

Answer 1

There is no easy way. You will need to group entire collection into array which may require allowDiskUse for large datasets with a huge performance impact.
db.collection.aggregate([
// count all documents
{ $group: {
_id: null,
cnt: { $sum: 1},
docs: { $push: "$$ROOT" }
} },
// add _batch field to group documents by
{ $project: {
_id: 0,
docs: { $map: {
// add a sequential number to each
input: { $zip: {
inputs: [ "$docs", { $range: [ 0, "$cnt" ] } ]
} },
as: "doc",
in: { $mergeObjects: [
{ $arrayElemAt: [ "$$doc", 0 ] },
// split it in batches by 4 based on the sequential number
{ _batch: { $cond: [
{ $eq: [ { $arrayElemAt: [ "$$doc", 1 ] }, 0 ] },
1,
{ $ceil: { $divide: [ { $arrayElemAt: [ "$$doc", 1 ] }, 4 ] } }
] } }
] }
} }
} },
{ $unwind: "$docs" },
{ $replaceRoot: { newRoot: "$docs" } },
// ensure original order, only if you need ItemRange as a string
{ $sort: { _id: 1 } },
// calculate averages per batch
{ $group: {
_id: "$_batch",
start: { $first: "$ItemName" }, // only if you need ItemRange as a string
end: { $last: "$ItemName" }, // only if you need ItemRange as a string
RatingAvg: {$avg: "$Rating"}
} },
// only if you need them in order
{ $sort: { _id: 1 } },
// calculate ItemRange, only if you need ItemRange as a string
{ $project: {
_id: 0,
ItemRange: { $concat: [ "$start", "-", "$end" ] },
RatingAvg: 1
} },
])
Not sure about practical usecase as all averages will change when you remove e.g. the first document.
Anyway if you don't need ItemRange in format "FirstName-LastName" and can live with batch number instead, you can skip 2 lasts in-memory sorts which should improve performance.